Optimizing Empirical and Hybrid Modeling for Advanced Canopy Chlorophyll and Nitrogen Retrieval Technique Using EnMAP Data

Abstract

• This study provides valuable insights into the performance of different modeling approaches for retrieving canopy chlorophyll content (CCC) and canopy nitrogen content (CNC) from hyperspectral imagery. • Model performance KRR is the best for CCC (empirical) and CNC (empirical, hybrid), GPR for CCC (hybrid) retrieval. • Band selection red edge to NIR for CCC, SWIR for CNC. • Nitrogen to protein conversion factor adjustment a significant improvement in CNC retrieval accuracy was achieved by adjusting the conversion factor from 4.43 to 3.03 when deriving CNC from leaf area index (LAI) and leaf protein content (Cp). • Comparison of approaches the research suggests that empirical techniques are well-suited for small agricultural areas, while hybrid models show promise for heterogeneous landscapes where ground truth data are limited. • EnMAP hyperspectral imagery potential EnMAP hyperspectral imagery shows potential for routine regional monitoring of CCC and CNC in agricultural areas. This study evaluates empirical and hybrid physical models for estimating canopy chlorophyll content (CCC) and canopy nitrogen content (CNC) using hyperspectral imagery from the Environmental Mapping and Analysis Program (EnMAP) over Michigan's Kellogg Biological Station in summer 2023. In the empirical approach, six machine learning regression algorithms (MLRAs) have been evaluated. In the hybrid modeling approach, each MLRA was combined with the PROSAIL radiative transfer model. Results show the empirical model outperforms the hybrid model for CNC retrieval, while both perform similarly for CCC. In the empirical approach, KRR demonstrated the best performance among MLRAs for both CCC (RMSE = 0.10 g/m², NRMSE = 9.76 %, R² = 0.93) and CNC (RMSE = 0.10 g/m², NRMSE = 8.13 %, R² = 0.94). In the hybrid modeling, GPR performed best for CCC (RMSE = 0.10 g/m², NRMSE = 9.43 %, R² = 0.93), while KRR remained the top performer for CNC (RMSE = 0.17 g/m², NRMSE = 13.67 %, R² = 0.83). Furthermore, the findings indicate that the hybrid model exhibits greater sensitivity in heterogeneous areas where field data are limited, while both approaches effectively distinguish between organic and non-organic treatments. The nitrogen conversion factor refined from 4.43 to 3.03 for corn in this study significantly improves the accuracy of the estimated CNC. This enhancement provides further evidence of the efficacy of EnMAP imagery in estimating biochemical parameters and its potential application in Precision Agriculture.

Topics & Concepts